Skip to content

How PoT-O Works

Mining Flow

1. Challenge derivation (from Solana slot hash)
         |
2. Tensor computation (AI3 engine)
         |
3. MML Path validation (compression optimality)
         |
4. Neural Path validation (activation matching)
         |
5. Proof construction + on-chain submission

1. Challenge Derivation

The latest Solana slot hash is read via RPC. From its bytes:

  • Byte 0 determines the operation type (matrix multiply, convolution, ReLU, sigmoid, tanh, dot product, normalize)
  • Byte 1 determines the tensor dimensions (clamped to max 64x64 for ESP compatibility)
  • Remaining bytes seed the input tensor values
  • Slot height influences difficulty scaling

This makes each challenge deterministic and verifiable.

2. Tensor Computation

The miner uses the AI3 engine to execute the selected tensor operation on the input tensor. Operations include:

  • Matrix self-multiplication
  • 1D convolution with a fixed kernel
  • Activation functions (ReLU, Sigmoid, Tanh)
  • Vector dot product and normalization

3. MML Path Validation

The Minimum Message Length score measures how well the output compresses relative to the input:

mml_score = compressed_size(output) / compressed_size(input)

A lower score means the transformation found a more compressible (optimal) encoding. The proof is valid only if:

mml_score <= base_threshold / (1 + log2(difficulty))

Higher difficulty requires tighter compression ratios.

4. Neural Path Validation

The tensor operation is modeled as a small feedforward network (default: 32 -> 16 -> 8 neurons). The "path" is the binary activation pattern at each layer (ReLU > 0 = 1, else 0).

From the challenge hash, an expected path signature is derived. The miner iterates a nonce parameter until the actual activation path matches the expected signature within a Hamming distance tolerance:

hamming_distance(actual_path, expected_path) <= max_distance(difficulty)

5. Proof Construction

The final proof contains:

  • challenge_hash -- the slot hash that seeded the challenge
  • tensor_result_hash -- SHA-256 of the output tensor
  • mml_score -- the compression ratio
  • path_signature -- hex-encoded activation bits
  • computation_nonce -- the nonce that satisfied both constraints
  • computation_hash -- SHA-256 binding all fields together

This proof is submitted to the tribewarez-pot-o Solana program for on-chain verification and reward distribution.

TribeWarez Blockchain Ecosystem